Anthropic says Claude 5 models need far less system prompting, urging developers to simplify agent context engineering

Anthropic says it removed over 80% of Claude Code’s system prompt for newer models like Claude Opus 5 and Claude Fable 5 without hurting coding benchmark results. The company argues developers should revisit context engineering across system prompts, Skills, CLAUDE.md files, memory, and other shared agent instructions.

The shift suggests frontier models may be becoming better at inferring intent with fewer guardrails, reducing the need for long, brittle prompt scaffolding. For teams building agents, the emerging best practice is likely minimal, high-signal context that avoids overconstraining model behavior while preserving necessary policies and workflow guidance.